Cancer Metaphors in Language: Corpus & Computational Analysis

[00:00:08]so as always we thank the fleshbook samuelson foundation and the school of social sciences humanities and arts it's my pleasure today to introduce juana david who got her uh undergraduate degree at bu in anthropology uh a master's degree in japanese studies and linguistics at oxford and a phd in linguistics from berkeley so a very nice interesting background she's doing very interesting work in constructed grammar construction grammar grammar um i'm gonna keep that little okay all right metaphor cognitive linguistics and very interesting stuff on things like data mining automatically identifying metaphors and texts this really cool meta-net project which we'll be talking about um so very exciting stuff and uh in for today's talk we're going to hold on questions until the end so with that thank you for inviting me to speak at this wonderful event and i've been really looking forward to presenting some of this work and hearing some feedback from or said folks and your ideas that i've gotten on and off throughout the last few weeks this is all going to come together now so i'm really excited um so i'm going to talk a little bit about some of the work i've been doing in collaboration with some folks including teenymatlog and including people at berkeley and at other universities and trying to formalize how we analyze uh metamorph and how we analyze that for discourse in a more computational way and in a more systematic way and in the more big data way um so we're moving sort of away from this line by line discourse analytic approach and integrating some of those insights with more computational approach um and this is all in the context of studying metaphors for cancer in this case because we have taken this turn now and looking at more applied ways of studying metaphor to see how people's world views and how their attitudes are shaped by the metaphors that they hear and the metaphors that they use so although this is a talk about cancer metaphors it's really a talk about metaphor theory in general methodologies and metaphor research in general so it's kind of like a masked metaphor theory talk um so just to go a little bit over the assumptions and the background of what we are talking about here um here are some of the insights that we take for granted from cognitive science and from linguistics we think that language influences thought and vice versa but also that these two things are influenced by action and they influence action so it's not just about sitting there thinking on your own it's about taking those thoughts and those feelings and those attitudes and using them to do things in the world like vote or create medicine or found a non-profit organization that helps people so all of this has real world applications and this is why we care about this kind of stuff in broader context we in cognitive science and kind of linguistics more generally we do not think of words as having inherent discrete meanings we see words as prompts to more broad meaning fields more broad meaning um frames as we call them semantic frames in the filmorean the chuck film more sense um so a word will act as a prompt to a broader set of meanings and you are not sort of processing words one by one uh you're prob you're processing them in the flow of discourse basically um and if we're talking about words as prompts to meanings words obviously appear inside of grammatical structures right we don't speak word by word we speak phrase by phrase sentence by sentence so we have to take into account the contribution of grammar to how we structure our meanings so this is where we get things like construction grammar or constructive grammar i don't know what to call it now because it is constructive um it's the study of how these words come together and create a compositionality of meanings that have their own unique structures independently of their separate components and of course as the topic of my talk suggests there's much metaphor conceptualization what is metaphor uh very very broadly speaking it's when you take some kind of less accessible meaning like an abstract concept like democracy or love or just the process of thinking and you think about it and talk about it in terms of a more accessible domain like motion through space object manipulation force dynamics with objects uh gravity verticality these are all sort of inter-subjectively accessible domains that we can both verify media speakers as a listener um so that's really the the backbone of the conceptual metaphor theory movement is to take this mapping from the source domain to the target domain as the basic metaphoric structure and that gives rise to all kinds of more complex metaphors later on so here's an outline of my talk first i'm going to talk a little bit about metaphor research more specifically computational and corpus linguistic approaches to medical research and research and then i'll talk a little bit about cancer discourse research and how cancer has been talked about in the domain of health communications and in the media and in rhetoric um and in the literature and in literature and poetic literature and then sort of put the two together and see how we can understand more about cancer and how it's being construed via these methods that i'll be presenting and by the technologies that we're developing and it's uh it's important to note that this is not unique to cancer talk this work has been done in the context of other societally important issues like poverty the gun debate and gun rights issues and democracy amongst many other topics so some of the principles i'll be presenting are more broadly applicable so what are the important questions an important question is not are there metaphors yes there are metaphors i just answered it at the end we can move on uh there are metaphors all the time even in my discourse right now as i'm speaking to you in my discourse this course is a container um i am using a lot of metaphor um so that's not what we're focusing on that's not what we're trying to answer we're trying to answer what metaphors are there and more precisely why those metaphors why is it that a particular person is using a particular metaphor at a particular time and are there generalizations we can draw from that in terms of the identity of the speaker and the audience and then from that we can ask what do speakers achieve by using particular metaphors it begs to be mentioned here we take it for granted that people don't consciously pick their metaphors most of the time most of the metaphors we use are at the unconscious level but sometimes we do get strategic especially if we're trying to be persuasive or if we're trying to frame a message in a particular way okay so um from the point of view of linguists and cognitive scientists how do we know what conceptual metaphors people are reasoning in terms of or even that they are using metaphor at all and there are three main approaches to studying this the first is looking at linguistic clues so this is the work of cognitive linguists and they look at lexical choices with grammatical choices and discourse level choices and this is mostly what i'll be talking about in what i'm presenting here but then there's the gestural approach where people look at co-speech gestures and see how gestures can uh kind of give an insight into some of the metaphors operating subconsciously that you're not explicitly saying in the words that your gestures are accompanying so a famous study of this which i'm sure you're all familiar with is the new nunez and sweets that work with imara time metaphors where they're going like this to indicate a long long time ago uh even though they're not saying a long long time ago but they're going like this this indicates that the past is actually in front and the future is behind so there yes there is language indicating a metaphor but the gesture is actually coming across with a lot of that information so these are some of the people that work on that and then of course there are non-linguistic or semi-linguistic tasks that try to get a metaphor without necessarily mediating through grammar through language um just to show that metaphor really is conceptual conceptual structure and not necessarily a linguistic or language specific structure okay so i'm going to be talking mostly about the linguistic clues um so if you're a linguistic student of metaphor what are you looking at what kind of linguistics are you doing um well you can be the armchair linguist and you can sit there and go i'm going to think of a metaphoric sentence now uh they were struggling in their relationship and they hit a dead end straight okay that sounds right that sounds about right to me right that's metaphoric i have heard people say this i'm going to use this as an example now to show that we reason about love in terms of journeys and in terms of streets and paths and destinations um we can take that one step further and do a more discourse analytic approach where we take actual naturally occurring speech conversations news articles and annotate line by line and say okay these are some of the metaphors i'm seeing here based on the words and phrases being used we can do even more intensive quantitative work um in a corpus linguistic tradition where we do concordances we do n-grams we do collocations and word frequencies and this is sort of zooming out from the immediate discourse context but giving you more data to work with and whenever you do that you lose precision but you gain threat so as i'm going through these i'm getting more and more broad and less and less precise and of course there are the statistical seed-based approaches which um they work to a certain degree of precision and the point is to get a better and better score so 0.7 this time 0.75 next time where you use particular collocations like cancer journey to seed additional discoveries of additional metaphors from particular collections of texts so what are some of the existing problems in metaphor research so far these have been not enough empirical evidence not enough cross-linguistic evidence not enough looking at metaphors and different domains of human reasoning and action which is why we are expanding to looking at cancer metaphors and democracy metaphors and property metaphors um and when we are looking at metaphors in these domains we have to take into account cultural models and we have to take into account political stance and world views and this becomes enormously complicated um that's where the discourse analytic approach is useful and all of this came up by the way earlier today in matt's talk so this is really tying in with research already being done here um a really big problem that um my research and the research of my uh group uh is trying to address is getting beyond a list-based model of medical we're not just listing metaphors one by one but we're creating structured networks of metaphors that are related in predictable ways and then we can discover new specific instances of metaphors based on this core set and of course even defining metaphors problematic what do we know when we know that an expression is made of work so here are some of the assumptions that that i take and other people take there's a conceptual metaphor that is grounded in so-called primary primary metaphors and primary metaphors are arising out of primary scenes so these are metaphors like progress is forward motion and purposes are destinations and difficulties in forward motion are impediments along the path to a destination another popular primary metaphor is affection is warmth and these are thought to be metaphors that arise in early infancy and in the developmental period of language when children are associating these different domains regularly and subconsciously and then these primary metaphors are the ones that give rise to more complex metaphors like cancer is a journey which is a more culturally understood metaphor so a child will not acquire cancer as a journey but cancer is a journey builds on the earlier metaphors that are primary and i think you earlier earlier you mentioned that i should talk a little bit about um the difference between metaphor and figurative language more broadly or blends for example so uh there are other forms of figurative language there are not conceptual metaphors in this definition um but those are not really what we're looking at when we are doing this kind of research so i can address this more in the q a session at the end just know that it's there metaphor is not just linguistic uh it points to a conceptual reality and some languages have these primary metaphors that i just mentioned but they might not necessarily manifest them linguistically so you go to a community that speaks a language that you don't know you try to look for the metaphors and you're not hearing them it doesn't mean that they're not there so you have to do the gestural research and you have to look at other ways to find them uh and the really important part of this two computational approaches to metaphor theory is that they're regular so you can actually formally encode them in a computational system they're not uh there's not going to be thousands of them there's actually only a few hundred of them and once you get those down pat you can leverage those to find more specifics so the basic model is uh when you when you're doing a little metaphor analysis uh you do your source domain frame like this for example physical combat with particular frame elements so you have at least the fighter and another fighter and some kind of outcome to the fight and then you have your target frame which is the the thing that you're trying to understand in terms of the source right so in this case you're trying to understand cancer as a fighter and a cancer patient is another fighter and the disease outcome survival or not is the outcome of that fight so somebody losing or winning so when doing this kind of research computationally there's always a trade-off between precision and breadth as i mentioned earlier so in a world where we have unlimited resources and i have 3 000 or 300 000 talented metaphor analysts and they work 24 hours a day uh they can generate extremely detailed highly precise analysis of metaphor we can compile all of those and let's say we generate corpora that are billions of words large and we're happy but we don't have those kinds of resources so we are forced to cut down on some some of the ideal that we would like and replace it with lower precision but higher breadth so some of the research groups that have done this kind of work but not with 300 000 analysts are the pragel jazz group which um has pioneered at the mpi has pioneered this famous um methodology and metaphor research that you have probably read about and in the context of cancer research it's been the gibbs and frank's work so ray gibbs has interviewed some women that talk about their cancer experience and done very very in-depth discourse analytic work over over their reports um a more preferred method for people who are interested in computationally leveraging this kind of stuff is one that uses some nlp methods so uh the seed-based methods that i've mentioned earlier that are mainly done by caterina who's at the university of cambridge right now i believe um these have very very wide breadth they produce hundreds of thousands of results but very low precision um and you're not really sure whether you're dealing with actual conceptual metaphor in the results or whether they're just very interesting complications of words or other forms of figurative language then there's what i call the winnowing approach uh which is really kind of cool uh you take um this collocational pattern analysis method where you search for a target word like cure or treat or ailment or disease or cancer and then based on the results of that you search that for source domain words like fight beat journey when war and then you find the collocations within that and you have your final list of potential metaphors this also requires a human analyst because the human analyst is the one who's going to go looking for this you still need someone who knows what they're looking at in terms of conceptual metaphor um but it's very highly focused on lexical items and it doesn't really get at the conceptual metaphors that i spoke about earlier so what are we really saying when we say fight cancer why fight cancer versus attack cancer what's the difference there there those are both sort of invoking this struggle physical struggle metaphor but in different ways so lexically they're different but conceptually they're sort of similar so we want to get conceptual metaphor in the mix and then finally um the fourth method that is uh is starting to arise now in the context of this project that jeff mentioned uh called metanet where we're trying to bring together the best of both worlds the qualitative and the quantitative approaches and we're trying to build a human analyst's intuitions into a computational system by creating an ontology of structures that are necessary for conceptual metaphors so for example in the system all 200 primary metaphors are there and they're interrelated to each other in a dynamic way there is a metaphor extraction mechanism that takes these metaphors and runs over very very large corpora discovering metaphoric collocations in texts and then the results of that produce metaphorically scored collocations with a greater or lesser degree of metaphoricity so it's not that it's either metaphoric or not which is difficult to decide right whether something is not work or not um it's to what extent does the system believe this is metaphoric on the scale of zero to one so once the human analyst receives this list of results the human analyst goes through and decides where there were false positives and in fact this is not metaphoric at all or whether maybe perhaps a collocation is particularly metaphoric like attack poverty maybe attack poverty got a 0.3 score in fact it should have gotten a 0.9 score because it is highly metaphoric and then the human analyst sort of creates a gold standard of examples feeds them back into the system and then refines the system so that the next time it runs it produces better results so that's the basic model of so overall we can characterize this system as one that has large scale data extraction capabilities it detects metaphoric language that increases inaccuracy over multiple iterations and crucially mirrors human-like intuitions about what should be metaphoric what should not be metaphoric and i'd like to add to this that it's completely adaptable such that if the state of the art changes and we decide that something else is actually metaphoric those new changes can be integrated into the system too so if somebody comes along and makes a really compelling argument that something is not metaphoric we can adapt the system so that it reflects the new insights from the literature okay so that's my um sort of spiel about the state of metaphor theory and medical research and um now i'm just going to talk a little bit about cancer research uh and and therefore in cancer so here's i'm just kick it off with a quote from susan sundock who wrote this wonderful book called illness as metaphor and ace and its metaphors a very long time ago i think it was 1972 but i could be making that up um she says illness is the night side of life a more honorous citizenship everyone who is born holds dual citizenship in the kingdom of well and in the kingdom of the sick although we all prefer to use only the good passport sooner or later each of us is obliged at least for a spell to identify ourselves as citizens of the other place and this was the first line of the book and then for the rest of the book she proceeded to try to convince us that we should not be using metaphor to talk about illness and to talk about cancer but clearly she preferred to use metaphor in the beginning and in fact really couldn't even avoid it and she did a very nice job creating a very novel metaphor and this is the cancer is a journey metaphor right this is the metaphor whereby you are a traveler with a passport and there are two states that you can be in healthy and not healthy and sometimes you're in the healthy state location and other times you're in the unhealthy state slash location and in fact research has shown that cancer is very frequently talked about as a journey in american english discourse and in british english discourse as well um and journey metaphors are more broadly a type of life is a journey cancer journey metaphors are more broadly a type of life is a drinking out of course where the state of having an illness like cancer is an event in your life and then you go from that event of any cancer to maybe being an event of not having cancer so you're sort of progressing through life and every subportion of the life isn't self is itself a mini journey that you go through and when uh we take apart the metaphor we actually find some mappings that are very regular so usually the diagnosis stage of the cancer journey is construed as the beginning of the journey the treatment phase is construed as the past so you have expressions like course of treatment right the course um and the end of life care is the end of the journey sometimes you have people saying the end of the journey to talk about the recovery stage um it's ambiguous in fact if you go through the blogs people say i've reached the end of my journey and you have a mixture of feedback at the bottom of the page going oh i'm so sorry just like no no i'm at the end of my journey i'm recovering now um so it could go either way sometimes the remission the survival stage is seen as overcoming some obstacle in the past and that obstacle is usually that really really hard part of the cancer suffering process doctors and loved ones are seen as guides and companions um and the illness itself is sometimes construed as the companions shouldn't it almost be the environment though usually so speakers get really creative with how they can screw it and sometimes they get contradictory in the same in the same discourse so earlier on they might say my love beloved companion in this journey referring to the loved one and then later they might say uh and this god-awful companion that i have in life meaning the cancer so yes people do mix metaphors that's a really interesting phenomenon too um so that's one side of the metaphor point in terms of cancer and the other side is we do use a lot of warm fighting metaphors and the fact most literature on metaphoric cancer is about the war metaphor so even if you just peruse through the amazon book selection on cancer self-help books and other specialized books um it's just full of war metaphors struggling battle et cetera et cetera now the the actual experimental work on metaphorical scroll and cancer discourse is very limited um this is one of the papers or series of papers is a seminar all which i recommended to some of you to read before this talk um and in this in this paper um they find that patients use both war and journey metaphors uh more than health care professionals uh but more so journey metaphors so um that's understandable right the healthcare professionals are probably using a lot of dry language a lot of neutral language and it makes sense that the patients would use more of both and it's also interesting that overall the war metaphors are preferred over the journey metaphors by either patients or doctors um so that's it's a very limited um series of results it's just the seminal papers this may not all papers and also this one that i wanted to mention the house of schwartz which is a couple of experimental studies they did showing that war metaphors are likely to cause healthy individuals uh to take fewer preventative measures or to not want to actively take preventative measures to stay above cancer um so it's not really looking at cancer patients it's looking at people who don't have cancer which kind of suggests that cancer is inevitable so what would you do to avoid it it's somewhere down the path and um it's forcing people to sort of stimulate what they would have to do to avoid that path and that's it this is the research on cancer um most of the other um most of the other um items in the literature are by oncologists uh science experts in the media who want to discuss current approaches to cancer treatment who have concerns and they express concerns about using the more metaphors too heavily so this opens a lot of a lot of questions a lot of doors for research first of all do different speakers use different metaphors at different times so even within cancer patients we can't just say cancer patients that's too broad of a category there are people who are just diagnosed there are people who have had cancer for many years and they've been living with it for many years there are people in remission and they talk about cancer in different ways they use different metaphors so i i really like this one a popular article by wendy harpum named um or called misguided metaphor in the arcology times 2007 she really points out that um that we really need to be paying attention to the identity of the speaker so if it's somebody who says um she says the battle metaphor doesn't work well for me as a patient but that's not the upsetting part it's this notion of losing so even a patient in the middle of struggling with cancer might find it discouraging to think about it as a war but it gets even worse if you're a patient who is terminally ill and you have absolutely no hope um then it really is counterproductive to think about it as a war so she says uh why do i get uh why do i get uh worked up about losing so why do i get worked up about losing her battle because in the context of end stage illness this metaphor is seriously flawed in two ways first no victor emerges from the battlefield of hospice um and then she talks about how there is harm that this metaphor does to patients in their final days and to their loved ones left behind um so again yes there are more metaphors but most likely not coming from these sets of patients and there's a very strong active push to not use these metaphors for these kinds of patients so another issue to really consider here is what precisely about cancer is being discussed um is it cancer as a personal problem or lifestyle or is it dealing with cancer at a societal level with policies and governmental governmental tactics these are very very different right if you're talking about your own cancer experience versus if you're talking about some law that was just introduced to make more funds available to cancer research or you can zoom in on the cancer patient him or herself and look at the cancer patient as a battleground for the cancer itself and talk about the cancer spreading through the body which is a whole other set of metaphors or the cancers eating at somebody so that's more of a bodily focus of the cancer it's not so much about the psychological struggle or the lifestyle or the path of life it's about the actual pain and the actual anatomy and the physiology of cancer so what do we mean when we say cancer is a war what are the mappings in these two different ways of looking at cancer as a war so i'll take this excerpt from what is this from scott i feel as though i have two enemies inside one is the disease one is the treatment that kills the tumor but i fear that it will kill me first and after that the tumor that's what i feel you have to be strong to handle both and that's a very very different way of talking about cancer as a battle than the war on cancer since 1971 when president nixon announced a war on cancer with the signing of the national cancer act we have seen an increase of over 300 000 in the number of survivors which is now reaching more than 12 million in the united states so there the mappings are a little bit different from here here cancer is seen as a societal problem which is the enemy of the members of that society the society members are the protagonists in the war so not the enemy the other person and the policies and society-level actions are the weapons in the war a completely different set of mappings from the tumor seen as the enemy of the patient and the enemy of the treatment because she's doing a control of battle between two different sets of enemies the treatment is the enemy anticipation or the enemy of the tumor and the patient is the protagonist but also acts as the battleground of the uh tumor fighting treatment um it is a genuinely mixed metaphor right there she is putting together two completely different sets of mappings in one so these are the kinds of questions we should be asking when we talk about cancer as a war what do we need which war okay so uh to try to answer some of these questions uh i'm looking at this using both traditional corpus linguistic methods which includes collocations frequencies and concordances and engrams and the line and matanet which i have just finished introducing and some of the hybrid techniques that are in a part of the internet i'm putting here and i'm sure that i can make my slides available to you and in any case i can make these links available to you we do have a public wiki site for the internet project on the project site and i put all my links on my site if you want to take a look um so i am using uh these sets of corpora which are called the 1010 web corpora the reason is number one they are extremely large these are the largest corporate that exists they were uh used to they were actually produced as a scraping of the internet on a series of days in that year so they just left the machine running for like a week and scraped the entire internet and then cleaned it up they compiled it using spiderlink they cleaned it they de-duplicated it so there are no duplicated links in the entire corpus and they used to tag her as as their tagger and more importantly for my interests they did this from multiple languages which is unbelievably useful um so and for all of them they have billions of words and in most studies more most corporate studies have metaphor you see things in the hundreds of thousands and in the millions of words and here we're talking about billions of words so we're getting really kind of good numbers in spite of there being billions of words we're getting very small numbers of hits for particular publications of interest for specific domains so here um this is where the relevance of grammar in the study of metaphor becomes important because when you're doing collocational research uh you're not just putting two random words together you are kind of interested how those words occur grammatically relative to each other so in the sketch engine tool that i use to describe these they actually do categorize these collocations in terms of the grammatical structures and here i'm comparing a verb plus direct object cancer so bowel cancer attack cancer in english with spanish and this is showing the raw frequencies this is the total frequencies but i haven't listed all of the words these are just the ones pertaining to fighting battling tackling all of the violence verbs um in english and in spanish so we have kamatir vancer and fentares extirpar avansara takar derotar destrir aborda some of the same words in both but with different frequencies these columns log dice this is a lexical association measure that shows how likely this particular collocation is to be to be to exist um so this is actually a better indicator of salience than the rock frequency because the rock frequencies as you can see are very small and they're very small relative to the total as well so um in in english we're seeing some very very large associations worse and in spanish maybe not as high but still kind of high but importantly when i scraped this for english i just took the um the battle metaphors and the war metaphors but those were not the highest scoring scoring metaphors there were journey metaphors that were scoring higher and other metaphors but for spanish these were the highest scoring metaphors and they're already scoring quite low or quite mediocre so uh that's kind of an interesting distinction right there between spanish and english going to show that it's important to do cross-linguistic work like this um also taking into account the grammatical patterns that we're looking at we can't just look at locations of words can i wait till the end okay sorry about that um okay then i did this this uh collocation for the effects of grammar in a um this is a popular construction so cancer is up complete the sentence and in english we have quite a quite a variety of cancer is uh cancer is a killer cancer is a gift a battle an enemy a journey a fight an a burden a mystery a um they're very very few so there were this many hits of cancer but there were only 18 of there were like less than less than 50 52 for most of them um killer is the most frequent one with 287 but importantly what we note here is that the association of war metaphors is much lower when in the popular construction than in the verb plus object construction so it really makes a difference what grammatical construction we're looking at if i wanted to research cancer metaphors in english and only looked at popular constructions i would get the conclusion that war metaphors are really not that important in english so it really makes a difference to categorize the data in terms of the grammar as well i also found these kind of cool other sort of marginal metaphors like cancer is a game of odds um which are very very few but it is possible to talk about cancer in other metaphoric ways that are not war or journey and these kinds of methods help us discover what some of those ways of talking about it are if we look at the lower frequencies so we can't just look at the top five okay now um when we look at this stuff cross linguistically warm metaphors are most salient in english um they are also salient in spanish although maybe not to the same extent um i also looked at japanese and just out of curiosity i looked at romanian which is another romance language like spanish because i thought oh surely surely it's going to be just like spanish no it is not i'm hard-pressed to find war metaphors in romanian we do not get things like war on cancer or to have a war with cancer we do see some of the fighting with as in a hand-to-hand uh lucha lupta as as we say to hit cancer i'm not sure that exists much but in japanese interestingly enough we do get a lot of very vivid direct hand-to-hand combat individual to individual not so much larger wars with strategies and weapons but a lot of physical combat between two people so we get a lot of hit cancer or slap cancer we have things like extinguish cancer do away with cancer things like that um in the journey domain um we do see a lot of life is a journey extensions uh in english uh and in japanese uh it's not so much the notion of journey as this long drown out process where you have a final destination and you maybe have a companion but it's more about the immediate motion process walking along a path and then finding obstacles along the path in your way so we get things like means to take take the cancer and make it fly so it's like launching it or making it go away from you um we talk about koeru which is like surpass um and mori which means ride as her past so you get this compound um verb with two different events and um the journey metaphors are just completely not present in romanian at least they are not they're not discoverable through the the term cancer so if i search for the word cancer there's no journey language surrounding the term cancer so these are some of the cross-linguistic differences that are arising um meaning or begging the question then are these patients reasoning about their cancer in different ways um because romanian speakers don't have the journey metaphors are they dealing with their answer in a different way are they using are they appealing to a different metaphor system or are they just suffering more because they don't have an outlet so these are some of the research questions that are then produced on the basis of this kind of work okay and now i'm going to talk a little bit about how um metanet and a hybrid system like that can help us do some of this work in a way that is both qualitative and quantitative can help scrape larger data sets so uh just to give an overview of the architecture of metanet it's a database of manually entered so-called schemas which are frames semantic frames that are that have a structure with roles so for example the journey frame would have a journey a path of journeying a destination a source of the journey and that would be the schema um and then so this is very similar to frame that if if any of you are familiar with grammar this is the same kind of structure um then uh we also have metaphors which are frame-to-frame mappings systematically related to each other in in the um in the wiki that we used and uh we did this for four languages so english is just one of them we also have persian russian and spanish specifically for mexico um and we have obviously a big discrepancy in numbers here um clearly it's a lot harder to work with persian data in fact one of the reasons we have less persian data is because of the internet being blocked so yeah that was a real problem but really more importantly with persian and russian it's a character issue um the the characters are really hard to work with and then uh we have 760 732 metaphors for english and 627 schemas and we i need to talk a little bit about how we transition from the theory of metaphor and how mappings work and how we include that in the system so we assume that lexical items and expressions evolve frames and in turn those frames populate the source domain and the target domain of the metaphor so this creates a really nice little connect a to b kind of puzzle and then those source and target domains actually occur in predictable patterns that are grammatical so now we are connecting lexical items to frames frames to metaphors and metaphors to grammatical patterns and this is our way into the corpus data this is how we are able to focus data so here's just a little visualization of how that would work but this is not a cancer metaphor i just took this from a different set of slides sorry but this is poverty as a disease poverty as a disease has a source frame which is disease a target frame which is poverty and then poverty is a disease is itself a um subcase of social problems or physical afflictions at a higher level of conceptualization other kinds of physical afflictions that are social problems include crime and drugs so you can see how this might produce subcase relations for other metaphors as well and then the lexical items are associated with the um the frame so poverty has the lexical items poverty impoverishment destitution indigence disease has disease illness sickness etc um and the lists are augmentable in the system so we can always add new words and then the frames themselves have sub case relations so poverty is a subcase of social problem which feeds this target domain up here and disease is this subcase a physical friction which feeds the source domain in the higher metaphor and everything is interrelated in this way in a very very massive network system these are some of the grammatical patterns that the system uses to find publication syntax it uses subject verb constructions like cancer strikes verb direct object like battle cancer verb prepositional phrase like move beyond cancer or battle with cancer the noun of noun construction onslaught of disease the possessive cancer's journey of war on cancer which is a noun prepositional phrase the very very common noun known construction which cancer journey is the most frequent for adjective noun hostile disease and copular cancer is a very aggressive opponent um and the grammar pattern list itself is augmentable so for example we do not have any passive constructions so destroyed by cancer would not be able to be found consistent but we can add it and it will be incorporated so as i mentioned there are predictable ways of matching uh metaphors to grammatical patterns so if you have a noun noun construction like cancer journey or cancer cure it's always the first noun that's going to evoke the target domain of the metaphor and it's always the second noun that's going to evoke the source feeling of the metaphor and that's that's the rule so we don't have to worry about finding any anything that doesn't match this pattern or missing something that doesn't match this pattern the construction patterns are necessary constraints uh i'm not a fourth language but are not sufficient constraints to guarantee metaphoricity so at the very least this is finding this very limited set of metaphors in this very limited set of grammatical patterns which is already quite quite a lot um and then we could always worry about adding more as needed and as as we discover them this is a snapshot of the metaphor wiki the meninet wiki and some of the ways that you can interact with them with them at wiki so um each this one is for a frame this is for the cancer frame specifically uh it's listed as a sub case of the disease frame uh the disease frame is live so you can click on it and go to the disease frame open it independently and then at the very bottom it shows you for what metaphors this frame is a target frame for and which metaphors it's a source frame for so in the system cancer as a frame is the target frame for cancer is a journey cancer is physical combat surviving cancer is harming cancer surviving cancer is winning the competition there are some um there are some frames that can be both source and target so for example we can use cancer as a source domain for some metaphors like for example um taxes are cancer taxes are a cancer spreading through the nation yeah exactly in which case that would appear here taxes are cancer um and then um as far as the entries for the metaphors go we have entries for the source frame which is physical combat target frame which is cancer we have a place where we can list the mappings and then the most important part of the metaphor entry is the part where we relate that metaphor to other metaphors in the network because that gives it its location its address in that network so here cancer is physical combat is related to cancer patient is physical combatant um and then i didn't take a screenshot lower down but the next one down would be um cancer is a sorry social problems are physical afflictions would be the next one up which would be the subcase relation that that would be most important in locating this metaphor in the rest of the network so when we look at things that way and we have this higher hierarchical relational system of metaphors and they're not just the list then all of a sudden it's not so outlandish why people are using war metaphors and journey metaphors we can actually contextualize these two very specific metaphors culturally specific metaphors in a broader network of more generalizable metaphors that could potentially exist in other languages so in this case cancer is a journey is related to life as a journey which itself is a sub a subcase of action is motion and progress is forward motion and that is at a very very high conceptually schematic level um and cancer is war is a subcase of cancerous physical combat which itself is cancerous physical competition which is action is an interaction with an entity so when you're having a visual competition you're you're not moving per se but you're interacting with some other entity out there in the world and the more specific metaphors that we see in the purpose data like the cancer monster or the cancer race is actually these data points are actually evoking metaphors that themselves are contextualizing this broader system and we see why cancer is a race is closer to cancerous war and not to cancer as a journey because there is they're in different neighborhoods in the relational network um it also opens up the possibility of discovering new and creative metaphors so if one day somebody decides to call cancer an octopus and the addressee of that statement understands instantly what is meant it's probably because cancer is an octopus is a subcase of cancer as a monster and and the octopus is a subtype of monster it's a it's a seed dwelling monster so here i just took a snapshot of um don't worry about reading the details here this is just to give you a neighborhood view of some of the action is motion progress is forward forward motion metaphor system um and this is just a very very small uh slice of it as you can see we have arrows going off to the sides and if we if we were actually in the wiki i could scroll back and forth um if i put the whole thing up it would cover the entire wall um but as you add more metaphors to this they are themselves linked up to the to the system and interactable so i can actually click on these and go to that metaphor and open up the entry um the next step in the in the metanet system is to take that database of metaphors and implement them in scraping corpora for for collocation metaphor collocation patterns we did a couple of pilot studies from cancer metaphors we used the globe the global web-based english corpus of american blogs and we also made a custom corpus scraped from three cancer patient forms um which is very small at 656 thousand words this one is probably a little bit better than globe because it's it's concentrated cancer data this is sure to be a very polarizer so what happens is um the uh the corpus data is pre-processed it's put through a limitizer attacker a dependency purser it's put through this entire mechanism this stuff says standard nlp practice and then on the other hand the um extraction mechanism is running the with the metanet wiki over this data and spitting out sentences that it thinks are that include metaphoric language to a certain degree of certainty and produces a so-called metaphoricity score so a really good metaphor c score would be anything over 0.7 and we found that 0.7 is actually pretty decent without a 4c score a 0.7 would be something like battle cancer attack cancer cancer destroys lives things like that and so um all of this all these corpora are fed into the the pre-processing the custom system produces the metaphoricity score um and then that links up to the conceptual metaphors in the metanet wiki and we get to see what the actual conception metaphors are relative to the linguistic collocations in the data uh when we did this for cancer data this these were the basic results these are binned by whether they pertain to physical combat more broadly competition more broadly journeys more broadly disease motion so this is like the cancer spreads or the cancer grows through your body and other whatever that may be um and in keeping with the results from the existing literature that i presented earlier we see that physical combat outweighs journey metaphors by a lot um and we took apart physical combat and competition because you could talk about winning the war uh winning uh sorry cancer won or can or i lost against cancer and not have it necessarily be about a actual fight so it could just be a competition broadly uh here are the same results but with the bars taken apart uh with a little bit more fine-grained semantic distinction amongst the lexical items so this is all of these words are pertaining to attacking all of these words are pertaining to physical combat although these words are about war these words are about burden so load the cancer load or be overburdened with cancer or the stress of cancer this is competition and this is journey so just from this it's very informative to see that there's very little there's very little lexical diversity in the journey plan of force and very much lexical diversity in the metaphors pertaining to wars and physical combat which is very informative um here are the results produced in terms of the metaphors themselves the conceptual metaphors and not the lexical items so over here we have war metaphors such as cancerous physical combat surviving cancer is harming cancer cancer treatment is war harm is physical injury we have one bar for competition which is surviving cancer is winning a competition these bars here pertain to motion so processes are movements diseases are burdens action is being in a location cancer is a journey end of an action is the end of a path and this one has to do with object manipulation so encouragement is physical support so that's that's all the support metaphors curing disease is stopping motion uh actually sorry i'm gonna re-categorize that one that one belongs over here because it has to do with motion but i think the reason it got um he got put with object manipulation is because it has to do with obstacles in the way and taking away obstacles and manipulating obstacles which is object manipulation um and um the way that we interact with the corpus data results once the extractor has done its scraping is through a program called mr viewer it was originally called the metaphor reposito the metaphor repository viewer mr viewer so we called it mr viewer like mr coffee um and in this form you can actually see the individual sentences with the source domain highlighted in red and the target domain highlighted in green and you can search over the data um using this form by lexical item or essential metaphor so these are just some of the results from mr viewer these um these examples have to do with actually this is just a mixed bag of examples this is about journeys and wars defeating cancer cancer journey cancer attacks etc then if you are really interested in one of the examples and you want to know more about it you can click on that example and it gives you a breakdown of some of the um syntactic and semantic properties of that sentence so it gives you the dependency parser visualized like this it gives you the source domain lexical item the source domain frame the target domain electrical item and frame and it gives you the score of that particular extraction instance which in this case is 0.951 for the sentence help me in the fight against cancer um and it gives you the grammatical construction that was used by the program to find that example okay so um these are just more examples of the same i won't go through them this one is a selection of examples that were found that pertain to a more focused view on the body so here we have uh catch the cancer at its earliest stages where it's like this entity that you that you catch or you can slow down the cancer or you can contain the cancer so in all these cases your body is seen as this container this this area where your cancer is occupying its spread can move and it's your job to to deal with that and slow it down um and then there are other source domains that if you run the extractor as with a wild card for source domain you can find some interesting things so if you don't limit it by saying i want journey and i want more metaphors and you say i want all kinds of anything you can give me the source don't need it'll find stuff like the shadow of cancer also that's a light metaphor where i guess negative would be dark and positive light so one of the pluses of this technique is that you can discover maybe new um new patterns that you weren't actually looking for if you know how to fine-tune the system and to look for uh wild cards okay so what are some of the um weaknesses or drawbacks or however you want to call them of the mennonite system um as opposed to say the 300 000 analysts doing discourse analysis for a perfect precision that's the ideal well in meta-net you only really find the metaphors you're looking for so you have to do the discourse analytic portion ahead of time to know what metaphors might be there and then build the metaphors into the system so that the extractor can find them once it's looking for them so at the very very beginning we actually used the literature and additional discourse analysis to find the metaphors and built them in and these are this is the inventory of metaphors currently in the system for cancer specifically um other problems are that um frames are not predefined as a source or a target for a metaphor so cancer is not there's no place in the system where it says cancer must be a source domain um the job of the grammatical construction is to determine whether this is a source or the target so because cancer can be both source and target as we saw earlier this might give some false positives like taxation is a cancer on society when you're actually looking for cancer is a burden on this patient another drawback is that you need incredibly large amounts of corpus scraping to get just a little bit of data and uh for the custom corpus that i showed you the seven hundred thousand word purpose we only got a thousand hits um and of the of the 1000 hits a great many of them were false positives so you just have to scale up higher and higher and higher if you want to get more and more data and um another drawback is um there's a limited detection of metaphor when target domain item is not present uh but it's not impossible so uh what do i mean by that well actually let me just show you the example sentences uh take sentences like this uh i just want to help in any way i can since we're all in this mysterious journey searching for the hope of light at the end of the tunnel and at no point in this sentence was the word cancer mentioned but we know they're talking about cancer and the only way that this sentence would be picked up usually is because the word cancer or something cancer related is present as well this one might have been picked up because something earlier in the discourse contained the word cancer but we just got lucky and we don't want things to be based on luck we want to be systematic about this so one of the next goals is to try to figure out how to deal with examples like this where the target domain is not explicitly mentioned but it's still very very potently metaphoric okay so um just to summarize and to look forward in this in this body of research um we really want to use this technology to address the big questions uh and those questions are do speakers differ in what metaphors they use and how much they even use metaphor at all we had discussions earlier today in maths talk we had discussions at the metaphor for the common good about how frequent our metaphors by women versus men how frequent are metaphors in the media versus regular citizens talking about their condition et cetera et cetera those those are really important questions and can we use this information to provide better treatment and to provide psychological counseling um would choice of words help sort of drive action in one way or another or a course of treatment in one way or another are there metaphors we're not taking advantage of we're not even on our radar the kind of system that i've detailed here could potentially be used to find novel metaphors that patients are getting creative with in their own verbal discourse on their condition and finally what about cross-linguistic variety what can we learn about human cognition by looking at these the same sets of metaphors across different languages or even just narrowing it down to the same target domain and leaving it open-ended is to try to discover creative ways in which people are talking about important issues like black cancer and poverty and immigration so that's that's the spiel um i would like to take this opportunity to thank very very many people that were part of this work uh elise tickles me sweetzer ellen dodge that's not luca i didn't have a picture of luca so i typed handsome italian men and i took the most handsome one that i could find so thank you luca who is the engineer who created mr viewer and many many other technologies in that net george lakoff joseph hong and kerry morman who's right here and uh also thank you tini matlock and lab who have given me so much feedback and i've contributed to this work over the years and xc and the metaphor for the common good that just happened recently and of course all that grand stuff is very important as well thank you so much